Revert "Revert "enable csgmv automatically on cuda"" (#14277)

docs/references/environment_variables.md

@@ -119,6 +119,7 @@ SGLang supports various environment variables that can be used to configure its
 | --- | --- | --- |
 | `SGLANG_WAIT_WEIGHTS_READY_TIMEOUT` | Timeout period for waiting on weights | `120` |
 | `SGLANG_DISABLE_OUTLINES_DISK_CACHE` | Disable Outlines disk cache | `true` |
+| `SGLANG_USE_CUSTOM_TRITON_KERNEL_CACHE` | Use SGLang's custom Triton kernel cache implementation for lower overheads (automatically enabled on CUDA) | `false` |
 
 ## Function Calling / Tool Use
 

python/sglang/srt/utils/common.py

@@ -3711,6 +3711,12 @@ def cached_triton_kernel(key_fn=None):
     """
 
     def decorator(fn):
+        # Auto-enable the custom kernel cache for CUDA, where it is
+        # known to be compatible.
+        if is_cuda() and not envs.SGLANG_USE_CUSTOM_TRITON_KERNEL_CACHE.is_set():
+            logger.debug("Detected platform CUDA, using custom triton kernel cache.")
+            return CachedKernel(fn, key_fn)
+
         if envs.SGLANG_USE_CUSTOM_TRITON_KERNEL_CACHE.get():
             logger.debug(
                 f"{envs.SGLANG_USE_CUSTOM_TRITON_KERNEL_CACHE.name} = True. Using custom triton kernel cache."